Editor's Pick: NVIDIA-powered data center for high-performance applications

BOXX introduces FLEXX, a new data center platform ideal for complex product development workflows, including rendering and deep learning. 

FLEXX is available in various sizes and features NVIDIA RTX GPUs, the new graphics compute platform that is changing how companies do complex tasks including engineering simulation, rendering and deep learning for artificial intelligence. 

BOXX says FLEXX can simultaneously support multiple types of compute nodes, both local and remote. NVIDIA Quadro virtual workstations can be provisioned in minutes, enabling designers and engineers working remotely to stay productive. 

“FLEXX provides all the power and performance of a desktop workstation inside a rack-mounted, high-density form factor,” says Bill Leasure, VP of Marketing at BOXX. “This unique system enables organizations to accelerate workflows and work remotely.” 

Nodes in FLEXX are measured as vertical units (VUs). The FLEXX chassis supports up to 10 1VU nodes or five 2VU nodes. Nodes can be mixed, and can be removed or inserted without disrupting neighboring nodes. The chassis is five rack units tall, and equipped with redundant power supplies. 

FLEXX is a new data center platform ideal for complex product development workflows. Image courtesy of BOXX.

Released simultaneously with the FLEXX is the RAXX P6G Jupiter, a 6U rack-mounted system, which can support up to 16 NVIDIA Quadro RTX 8000 GPUs. Predictive Science, Inc. (PSI) recently completed a NASA-funded project to add initial GPU-acceleration to PSI’s magnetohydrodynamic solar physics simulation code called MAS. The PSI team used OpenACC to ensure portability and a single code base. One key application area of MAS is space weather, where it allows scientists to simulate multiple, parametrized models of solar eruptions to try to match observations that could eventually lead to a real-time predictive system.  

“Many simplified simulations are used to refine the parameter choices,” said Dr. Ronald M. Caplan, PSI in a press statement. “While these simulations can run quickly on a large HPC cluster, it is highly desirable to run them in-house on a single system to avoid wait queues and achieve a quick turnaround. Running a 60-million cell simulation on a 32-core processor takes nearly 3.5 days to complete; however, by utilizing all 16 NVIDIA Quadro 8000 GPUs in the RAXX P6G Jupiter, the same simulation runs in only one-and-a-half hours.”  

BOXX is now taking orders for both the FLEXX and RAXX systems. 

More info about the FLEXX can be found here.

For blogs on BOXX’s various technology offerings, click here.

Sources: Press materials received from the company and additional information gleaned from the company’s website.